Dynamic associations between cannabis use and sleep in adolescents and young adults during a cannabis intervention trial

Abstract

Improving cannabis treatment for adolescents and young adults (AYA) is a public health priority. Sleep difficulties may serve as a treatment barrier, as AYA may use cannabis as a sleep aid and cessation may induce withdrawal-related sleep problems. While research has identified associations between cannabis use, CUD, and sleep, few studies have examined these associations during AYA treatment, and no studies have conducted day-level analyses. The present study examined day-level, temporal associations between cannabis use and sleep difficulties during AYA CUD treatment. From 2009 to 2012, AYA (N = 65, 51% female, 15–24 years, 57% White) completed a 42-day ecological momentary assessment study while receiving cognitive behavioral therapy plus motivational enhancement therapy. Each day, participants reported on cannabis use quantity, sleep duration, and trouble sleeping. We used time-varying effect modeling to examine how day-level associations between cannabis use, sleep duration, and trouble sleeping changed across treatment, and if CUD severity moderated these associations. During the first week of treatment, cannabis grams were related to longer sleep among AYA with severe CUD and shorter sleep among AYA with mild CUD. During the second week, greater cannabis grams related to shorter sleep duration, regardless of CUD severity. Additionally, during these first 2 weeks, cannabis grams were related to reduced trouble sleeping. Cannabis use was otherwise unassociated with sleep duration and trouble. Findings suggest clinicians treating AYA CUD should provide greater sleep management skills early in treatment.

Cannabis is widely used among adolescents (ages 12–17) and young adults (ages 18–25; Johnston et al., 2023; Patrick et al., 2024). Early or frequent cannabis use is associated with myriad adverse outcomes including impairments in memory, motor coordination, judgment, brain development, educational and vocational achievement, and life satisfaction, highlighting an urgent public health concern (National Academies of Sciences, 2017; Volkow et al., 2014). These risks are most notable among those who initiate cannabis use during adolescence (Volkow et al., 2014). Although evidence-based behavioral interventions (e.g., cognitive behavioral therapy, motivational enhancement therapy) reduce cannabis use and related harms among AYA (Bender et al., 2011; Squeglia et al., 2019), few young people sustain these outcomes and even fewer achieve abstinence. Emerging pharmacotherapies also show promise; however, findings among AYA are mixed (Gray, 2013; Miranda et al., 2017). Elucidating factors impeding intervention effectiveness is necessary for improving treatment outcomes and advancing clinical care.

Sleep is fundamental to health, especially for AYA, and mounting evidence implicates its relevance for cannabis use. Adolescence and young adulthood are associated with well-documented neurodevelopmental shifts in sleep patterns. During this critical developmental window, youth often go to bed later and sleep longer; however, this can conflict with other life responsibilities (e.g., early school start times, balancing education and work obligations; Crowley et al., 2018; Teixeira et al., 2006). In turn, many AYA face sleep difficulties (e.g., short sleep duration, trouble sleeping; Hysing et al., 2013; Ohayon et al., 2000). While there are differences in adolescent versus young adult sleep patterns (e.g., sleep and wake times, sleep duration), the prevalence of sleep disorders is comparable across these groups, with nearly 20% of AYA meeting diagnostic criteria for a sleep disorder (Hysing et al., 2013; Ohayon et al., 2000). Although some AYA use cannabis as a sleep aid (e.g., young adults reporting quicker sleep onset time), it also impairs overall sleep quality (e.g., REM disruptions, reduced efficacy; Goodhines et al., 2019; Winiger et al., 2021). Over time, cannabis use can adversely alter AYA sleep patterns and lead to poorer sleep health (Hatoum et al., 2022; Pasch et al., 2012). Nonetheless, continued use is associated with increased expectations that cannabis use improves sleep among young adults and adults (Winiger et al., 2021), and using to relieve sleep-related withdrawal symptoms (Cohen-Zion et al., 2009) may reinforce sleep-related expectancies among AYA.

This link between cannabis use and sleep is heightened among AYA with cannabis use disorder (CUD; Cornelius et al., 2008), and studies are beginning to show that it can affect treatment outcomes. AYA with CUD are especially prone to sleep-related difficulties, which often occur during cannabis withdrawal and worsen during cannabis treatment (Cornelius et al., 2008) and cessation attempts (Cohen-Zion et al., 2009). These sleep difficulties typically begin within 24 h of cessation, peak within the first week of abstinence, and may take up to a month to resolve (Cohen-Zion et al., 2009; Sullivan et al., 2022). Concerningly, few AYA receiving cannabis treatment achieve sustained abstinence (Bender et al., 2011), often experiencing repeated intermittent periods of cannabis cessation that may prolong sleep difficulties. Indeed, sleep difficulties contribute to poorer AYA and adult treatment outcomes (e.g., less use reduction, shorter time to relapse; Babson et al., 2013; Parnes et al., 2023).

Unfortunately, despite high clinical relevance, whether and how day-to-day changes in cannabis use affect AYA sleep during treatment is largely unknown. Research examining the link between AYA cannabis use and sleep has relied heavily on nonclinical samples and cross-sectional and human laboratory data poorly suited for testing within-day effects of cannabis use on sleep and exploring whether they change over the course of treatment (e.g., Cohen-Zion et al., 2009; Vandrey et al., 2005). Examining trends in day-level, temporal associations between AYA cannabis use and sleep during outpatient treatment is a crucial step towards understanding whether reduced cannabis use impacts sleep and, if so, for how long. This information is essential for modifying interventions to improve sleep-related support at key points during treatment.

This secondary data analysis leveraged 42 days of ecological momentary assessment (EMA) data collected from treatment-seeking AYA (aged 15–24) in the context of a clinical trial studying the effects of topiramate plus a motivational enhancement and cognitive behavioral therapy (MET+CBT) behavioral intervention platform. AYA were randomly assigned to either topiramate or placebo, and all youth received the MET+CBT intervention. Topiramate is a sulfamate-substituted fructopyranose derivative that reduces alcohol, cocaine, and nicotine use in clinical trials (Johnson et al., 2003, 2013; Kranzler et al., 2014; Miranda et al., 2016; Miranda Jr. et al., 2008; Oncken et al., 2014). It has multiple mechanisms of action, including blockade of voltage-sensitive sodium and calcium channels, potentiation of γ-aminobutyric acid (GABA), enhancement of GABA_A receptor function, antagonism of AMPA/kainate glutamate receptors, and inhibition of carbonic anhydrase (Shank et al., 2000; Simeone et al., 2006). Studies show it reduces cannabis use in youth (Farrow et al., 2024; Miranda et al., 2017), possibly by modulating cannabis-related GABA and glutamate receptor activity (Moreira & Lutz, 2008).

For the psychosocial platform delivered to all youth, therapists used MET techniques to collaboratively set treatment goals with participants, which ranged from considering use reduction to abstinence. Prior research from this study found that all AYA reduced their cannabis use frequency, and AYA receiving topiramate also reduced their cannabis use quantity during use episodes compared to placebo (Miranda et al., 2017). Additionally, prior topiramate studies with AYA and adults found no associations between topiramate, daytime sleepiness, trouble sleeping, or sleep architecture (Jain & Glauser, 2014; Miranda et al., 2016), so we do not anticipate that topiramate use would impact AYA sleep in this study.

We examined day-level, temporal associations between cannabis use, sleep duration, and subjective trouble sleeping across the treatment trial, and tested whether the magnitude of these associations varied by CUD severity. Since AYA withdrawal-related sleep difficulties onset within 24 h, are most prominent early in treatment, and lessen over time (Cohen-Zion et al., 2009), we hypothesized that (1a) youth would experience shorter sleep duration and more trouble sleeping on nights when they did not use cannabis that same day, and the strength of these associations would weaken across treatment, and (1b) these associations would be stronger among youth with more severe CUD, as CUD severity is related to withdrawal symptom severity. While using cannabis may alleviate withdrawal symptoms, higher doses of cannabis may still induce sleep impairment among young adults and adults (Velzeboer et al., 2022). In turn, we hypothesized that (2a) on cannabis use days, using more cannabis grams would be associated with shorter sleep duration and more trouble sleeping that same night, and the strength of these associations would increase across treatment (as youth reduce use/tolerance). Given that individuals build tolerance to sleep-related effects (Babson et al., 2017), we also hypothesized that (2b) CUD severity would moderate these associations such that greater CUD severity (thus greater tolerance) would weaken associations between cannabis grams, shorter sleep duration, and trouble sleeping. Lastly, since alcohol use can also impact sleep, all analyses controlled for sameday alcohol use.

METHOD

Participants and procedures

The study involved AYA, aged 15 to 24, recruited from community settings (e.g., recreational areas, high schools). Study enrollment occurred from 2009 to 2012. Eligible participants had a specific goal of seeking assistance to reduce their cannabis use. Individuals mandated to undergo treatment due to court or parental orders were excluded. All participants reported using cannabis at least twice weekly over the previous month and demonstrated significant issues related to their cannabis use, with 80% meeting the DSM-IV-TR criteria for cannabis abuse or dependence. Importantly, participants were not required to have an official diagnosis of CUD since the DSM-IV-TR did not fully address substance use disorders in younger populations.

Exclusions were made for those who had received cannabis treatment in the last 30 days, currently experienced nonsubstance-related Axis I psychopathology (per DSM-IV-TR), or reported current suicidality or psychotic symptoms. Individuals with medical conditions or taking medications that would conflict with topiramate use or, in the case of females, were pregnant, nursing, or unable to use birth control were also excluded. A study physician assessed medical eligibility through an extensive review of medical history, physical examination, and laboratory tests.

This research was part of a larger clinical trial registered at http://clinicaltrials.gov (NCT01110434). Potential participants were screened over the phone to determine whether they met eligibility criteria, followed by a comprehensive in-person interview to confirm their status. Written informed consent was obtained from participants aged 18 to 24, while minors provided assent, and their parents provided permission. This study took place in the Northeast United States and participants were recruited from the surrounding region. At the time of this study, cannabis was medicinally legal in the study's state, but illegal in all forms in the surrounding states. This study was approved by the author's Institutional Review Board.

Enrolled participants completed a baseline assessment and training on how to accurately estimate cannabis grams by weighing a surrogate substance (oregano). This method of estimating daily quantities of cannabis use produces reliable outcome data (Norberg et al., 2012). During the baseline monitoring period, which occurred for approximately 1 week prior to randomization, and during the 42-day clinical trial, participants provided EMA data multiple times daily (Figure 1). They completed assessments in their usual environments using handheld wireless devices (Omnia; Samsung Electronics, Ridgefield Park, NJ) and custom software. The EMA program used simple English to present instructions, and participants recorded their responses directly in the program. They reported their data several times daily to capture a wide range of relevant variables. This secondary data analysis focused on morning reports, completed once-daily shortly after waking. Participants received $10 daily for complying with the EMA protocol.

FIGURE 1.

Study assessment timeline. Ecological momentary assessment (EMA), cannabis use disorder (CUD), timeline followback (TLFB), motivational enhancement therapy plus cognitive behavioral therapy (MET+CBT).

Study interventions

After participants completed baseline assessments, they were randomly assigned to one of two 6-week treatment groups: topiramate plus MET+CBT or a placebo plus MET+CBT. An investigator not in direct contact with participants utilized a computer-generated random allocation sequence to assign them in a 2:1 ratio (topiramate to placebo). The study employed a randomized block design with block sizes of 8, stratifying participants by sex, cannabis dependence, and baseline working memory performance (measured by Memory for Words) to ensure balanced treatment conditions. This imbalanced randomization strategy was designed to ensure an adequate number of participants in the topiramate group reached the target dosage, accounting for expected drop-out rates (Johnson et al., 2003).

MET+CBT

All participants, regardless of medication condition, received three 50-minute MET+CBT sessions, delivered individually during the first, third, and fifth weeks of the intervention. The MET+CBT protocol was based on previous research (Walker et al., 2006) and utilized principles of Motivational Interviewing (MI; Miller & Rollnick, 2012). The initial session aimed to enhance motivation for reducing or quitting cannabis use by building rapport, assessing readiness for change, and discussing the perceived advantages and disadvantages of cannabis use. In the second session, participants received personalized feedback about their cannabis use, which was discussed in an MI-style conversation. The third session involved a review of previous sessions, discussions about progress toward established goals, exploration of barriers to change, and assistance in problem-solving to set new goals. While sleep was not explicitly targeted, if participants reported sleep problems as a treatment barrier, they were provided with brief sleep hygiene psychoeducation.

To maintain consistency across treatment conditions, MET+CBT was administered by two master's and three doctoral-level counselors, ensuring that individual counselor traits or styles did not influence variations between conditions. All counselors underwent training of at least 20 h in MET+CBT, reviewed the MET+CBT manual, and conducted at least two mock cases. The same counselor facilitated all three sessions for each participant to foster rapport, continuity of care, and minimize attrition. Counselors were blinded to participants' medication conditions and did not handle any research assessments.

Medication condition

Participants and study staff who interacted with them were unaware of treatment assignments. An independent compounding pharmacy provided identical topiramate and placebo capsules. Participants randomized to topiramate underwent a 4-week titration to the 200 mg/day target dose and stabilized at that dose for 2 weeks. Topiramate capsules contained the active medication in specified unit dosages, while placebo capsules contained an inert filler.

Measures

Time-varying measures

Substance Use.

Prior-day cannabis use was assessed during morning reports by asking participants, “How many grams of pot did you smoke yesterday?” to the nearest 1/10th of a gram, accounting for the number of people they shared it with. Prior-day cannabis nonuse was coded as 0 = any cannabis use yesterday or 1 = no cannabis use yesterday. Considering that alcohol use can impact sleep, alcohol use was assessed as a covariate. Prior-day alcohol use was assessed during morning reports by asking participants how many standard drinks of beer, wine, or liquor they consumed, which we coded as 0 = no alcohol use yesterday or 1 = any alcohol use yesterday. We used the completion date for morning reports to define if the prior day (i.e., day when potential substance use occurred and sleep episode began) was a weekday (0 = Monday and Friday) or a weekend (1 = Saturday to Sunday).

Sleep Indicators.

During the EMA study, participants were sleeping in their natural environments (e.g., home, dorm). Participants were directed to activate “sleep mode” on the EMA program each night before going to sleep, which then prompted participants to set an alarm clock for their preferred waking time the following morning. Each morning when the alarm sounded, participants could either confirm they were awake or “snooze” the alarm for 10-minute increments. To calculate sleep duration, and consistent with prior research (Parnes et al., 2023), we measured the number of hours and minutes in between when the participant activated sleep mode at night and confirmed waking the following morning (e.g., after snoozing). To measure sleep trouble, morning reports asked participants, “How much trouble did you have sleeping last night?” with response options on a visual analog scale from 0 (none at all) to 10 (extreme trouble).

Static measures

Participant demographics (i.e., age, sex, race, ethnicity) were assessed at baseline. CUD was measured at baseline using the Kiddie Schedule for Affective Disorders for School-Aged Children for DSM-IV-TR (Kaufman & Schweder, 2004), which assessed 11 symptoms of CUD. We summed the number of endorsed CUD symptoms to define a continuous measure of CUD severity. Also at baseline, participants completed a 90-day Timeline Followback (TLFB; Sobell & Sobell, 1992), which we used to calculate a sum of the number of baseline cannabis use days. To account for individual differences in baseline sleep indicators, we used data from the baseline monitoring period to calculate average sleep trouble and average sleep duration at baseline (using the same measures as the time-varying measures). Lastly, participants were either randomized to receive topiramate (1) or a placebo (0).

Data analysis

First, we used R 4.4.0/RStudio 2024.04.0 (R. Core Team, 2023) and the “tidyverse” package (Wickham et al., 2019) for data wrangling and the “performance” package (Lüdecke et al., 2021) to calculate intraclass correlations (ICC), which represent the ratio of between- and within-person variance. We retained participants who contributed at least 2 days of EMA. Next, we used time-varying effect modeling (TVEM) using the SAS TVEM Macro (Li et al., 2015) in SAS 9.4 to examine day-level, temporal associations between cannabis use and sleep. TVEM can leverage multilevel data (i.e., days [Level 1] nested in persons [Level 2]) to examine associations between a predictor (e.g., cannabis nonuse) and outcome (e.g., sleep duration) over time (Shiyko et al., 2014). TVEM estimates regression coefficients for each unit of time (e.g., days), then generates a nonparametric spline function to estimate the direction and magnitude of coefficients over time. TVEM is appropriate for treatment outcome research, as many behavior changes occur with nonlinear, temporal patterns (Hallgren et al., 2018; Witkiewitz et al., 2022). TVEM results are shown as graphic depictions of estimated regression coefficients (with 95% confidence intervals) over time. Significance was established by coefficient 95% confidence intervals that did not contain 0.

We used linear TVEM to model sleep duration and Poisson TVEM to model sleep trouble, as sleep trouble was a positively skewed integer variable. Prior-day cannabis grams were person-centered (i.e., centered on each individual's average grams) and age, baseline sleep duration, baseline sleep trouble, CUD symptoms, and baseline cannabis use were grand-mean centered. We also included a grand-mean centered variable for cannabis nonuse and cannabis grams to parse apart within- and between-participant variance for these constructs (e.g., overall proportion of cannabis nonuse days from cannabis nonuse on a given day). When estimating models, we first estimated intercept-only models for each outcome (i.e., two models) to examine day-level trends in sleep duration and trouble. Next, we added focal predictors, cross-level interaction terms, and covariates to the model. Lastly, we removed any nonsignificant interaction terms from the model (of note, results in all models were consistent with and without inclusion of the interaction terms).

We estimated four adjusted TVEMs: cannabis nonuse, CUD severity, and the nonuse × CUD interaction predicting (Model 1) sleep duration and (Model 2) trouble sleeping; and grams on use days, CUD severity, and the gram × CUD interaction predicting (Model 3) sleep duration and (Model 4) trouble sleeping. Due to our focus on treatment-related changes and associated hypotheses, we only examined cross-level interactions (i.e., day-level cannabis use/grams with CUD severity). Models controlled for baseline sleep duration or trouble, respectively, age, baseline cannabis use, prior-day alcohol use, weekends, and medication condition. Baseline covariates controlled for behavior prior to the intervention, which allowed us to better examine treatment-related effects.

Parameter estimates from Poisson TVEM models were exponentiated to calculate incidence rate ratios (IRR), which describe the percent change in the outcome variable per one unit change in the predictor variable (Hilbe, 2014). IRR 95% confidence intervals containing 1 were considered nonsignificant. We probed significant interactions by plotting estimated sleep duration or trouble sleeping using the CUD mean (4.35 symptoms) and ±1 standard deviation (2.16 symptoms), which corresponds with DSM-5-TR mild, moderate, and severe CUD (American Psychiatric Association, 2022).

RESULTS

Participant characteristics and EMA compliance

Of the 66 randomized participants, we removed one due to missing data for all focal outcomes, leaving 65 in our analytic sample (N = 39 topiramate, N = 26 placebo). Our analytic sample was 50.8% female, M_age = 19.72 (SD = 2.18), and 56.9% White (Table 1). The mean number of endorsed CUD symptoms was 4.35 (SD = 2.16, range 0–9), and mean baseline cannabis use was 62.2 days in the past 90 days (SD = 25.96).

Table 1.

Participant demographic characteristics.

	Full analytic sample (n = 65)	Topiramate analytic sample (n = 39)	Placebo analytic sample (n = 26)
Variable	M (SD)	M (SD)	M (SD)
Age	19.72 (2.18)	20.31 (2.04)	18.85 (2.11)
CUD Symptoms	4.35 (2.16)	4.13 (2.07)	4.69 (2.29)
	N (%)	N (%)	N (%)
Sex
Female	33 (50.8%)	19 (48.7%)	14 (53.8%)
Male	32 (49.2%)	20 (51.3%)	12 (46.2%)
Race
White	37 (61.7%)	22 (59.5%)	15 (65.2%)
Black/African American	17 (28.3%)	10 (27.0%)	7 (30.4%)
Native Hawaiian/Pacific Islander	1 (1.7%)	1 (2.7%)	0 (0.0%)
Asian	2 (3.3%)	1 (2.7%)	1 (4.3%)
American Indian/Alaska Native	3 (5.0%)	3 (8.1%)	0 (0.0%)
Missing	5 (7.7%)	2 (0.5%)	3 (11.5%)
Hispanic/Latin	14 (21.5%)	7 (17.9%)	7 (26.9%)
Grade
9	1 (2.6%)	1 (4.5%)	0 (0.0%)
10	1 (2.6%)	0 (0.0%)	1 (5.9%)
11	3 (7.7%)	2 (9.1%)	1 (5.9%)
12	7 (17.9%)	3 (13.6%)	4 (23.5%)
13	7 (17.9%)	3 (13.6%)	4 (23.5%)
14	9 (23.1%)	6 (27.3%)	3 (17.6%)
15	4 (10.3%)	3 (13.6%)	1 (5.9%)
16	7 (17.9%)	4 (18.2%)	3 (17.6%)
Not in school	26 (40.0%)	17 (43.6%)	9 (34.6%)

Note: Cannabis use disorder (CUD). All participants received the Motivational Enhancement Therapy plus Cognitive Behavioral Therapy intervention. Grades 13–16 refer to college/university.

During the baseline monitoring period, the mean number of completed EMA days was 6.33 days (SD = 1.33, range 4–13, 98.6% compliance, 412 total morning reports across participants). On these days, the mean hours slept per night was 9.69 h (SD = 1.43), and participants endorsed any sleep trouble on 51.7% of nights (M_trouble = 1.82, SD = 2.58). During the EMA 42-day clinical trial, the mean number of completed morning reports was 31.89 (SD = 13.19, range 2–42, 75.9% compliance, 2033 total morning reports across 65 participants). Sex, age, CUD severity, baseline sleep duration, and baseline trouble sleeping were uncorrelated, while topiramate (r = .25, p = .048) and the number of EMA cannabis use days (r = .55, p < .001) were significantly correlated with the number of missing morning reports per participant.

While trouble sleeping data were available on morning reports in 2033, 154 corresponding bedtimes were missing, which left 1879 EMA treatment days from 65 participants (M = 28.91 days, SD = 13.20, range 1–42) available for analyses predicting sleep duration. The mean number of hours slept per night was 9.80 h (SD = 2.31, range 2.36–20.37). Participants endorsed any trouble sleeping on 47.7% of treatment days (M_trouble = 1.65, SD = 2.47). ICCs noted that 33% of the variance in sleep duration and 54% of the variance in trouble sleeping was attributed to between-person factors. Participants endorsed cannabis use on 59.1% (1201) of the 2033 total treatment days (M = 62.3% use days, SD = 27.9%, range 2.3%–100%) and on 58.2% of days with sleep duration data (1879 days total). On the 1201 use days, the mean number of cannabis grams consumed was 1.32 grams (SD = 1.16).

TVEM results

Cannabis nonuse

Cannabis nonuse models examined all study days. Intercept-only TVEMs (Figure 2) found that estimated sleep duration remained stable across the 6-week treatment period, with an estimated average of 9.83 h slept (estimated hours per day range 9.61–10.15). The intercept-only TVEM for trouble sleeping found that estimated trouble sleeping had a mild decrease across the treatment period from 1.89 to 1.56 (minimum 1.46).

FIGURE 2.

Intercept-only TVEM estimating sleep duration and trouble sleeping. Black lines reflect all treatment days and gray lines reflect cannabis use days. Solid lines depict estimated values and dotted lines depict 95% confidence intervals.

In the adjusted model predicting sleep duration (Model 1), cannabis nonuse (Levels 1 and 2), CUD, the cross-level interaction, and topiramate were all unrelated to sleep duration. Prior-day alcohol use was associated with shorter sleep duration from days 4 to 23 (M_b = −0.80, b range −0.58 to −0.87) and 33 to 40 (M_b = −0.78, b range −0.67 to −0.94), while weekends (M_b = 0.64, b range 0.52 to 0.94), baseline sleep duration, and baseline cannabis use were associated with longer sleep duration (Figure 3, Table 2). In the adjusted model predicting trouble sleeping (Model 2), only baseline sleep trouble was associated with sleep trouble (IRR = 1.31, p < .001); cannabis nonuse (Levels 1 and 2), CUD severity, the cross-level interaction, alcohol use, weekends, baseline cannabis use, and topiramate were all unassociated with sleep trouble (Figure 3, Table 2).

FIGURE 3.

Time-varying adjusted effects of cannabis non-use on sleep duration (top) and trouble sleeping (bottom). Models 1 and 2 (see Method). Black lines indicate regions where the 95% confidence interval did not contain 0 (significant), while gray lines indicate regions where the 95% confidence interval did contain 0 (non-significant). See Table 2 for level 2 results from this model.

TABLE 2.

Time-invariant (Level 2) TVEM results.

Variable	b	SE	p
Sleep duration on all days (1879 days)
Age	−0.05	0.07	0.49
CUD severity	0.01	0.08	0.91
Cannabis nonuse	1.08	0.65	0.09
Topiramate	−0.23	0.29	0.43
Baseline sleep duration	0.41	0.10	<.001
Baseline cannabis use	0.02	0.01	0.01
Sleep duration on use days (1093 days)
Age	0.06	0.07	0.39
CUD severity	−0.02	0.09	0.82
Cannabis grams	0.19	0.18	0.31
Topiramate	−0.29	0.33	0.37
Baseline sleep duration	0.42	0.09	<.001
Baseline cannabis use	0.01	0.01	0.04
Variable	IRR	SE	p
Sleep trouble on all days (2033 days)
Age	1.05	0.04	0.21
CUD severity	1.06	0.04	0.14
Cannabis nonuse	1.06	0.39	0.89
Topiramate	1.06	0.17	0.73
Baseline sleep trouble	1.31	0.06	<.001
Baseline cannabis use	1.002	0.01	0.65
Sleep trouble on use days (1201 days)
Age	1.09	0.05	0.06
CUD severity	1.10	0.05	0.04
Cannabis grams	0.98	0.10	0.82
Topiramate	0.91	0.19	0.64
Baseline sleep trouble	1.38	0.05	<.001
Baseline cannabis use	1.001	0.004	0.75

Note: Beta is presented for linear models (sleep duration) and IRR is presented for Poisson models (sleep trouble). Significant results are bolded.

Abbreviations: b, beta; CUD, cannabis use disorder; IRR, incidence rate ratio.

Grams on use days

Cannabis grams models examined only cannabis use days. Intercept-only TVEMs (Figure 2) found the average estimated sleep duration on use days was 9.73 h (estimated hours per day range 9.48–10.11), and estimated trouble sleeping on use days decreased from an average of 1.90 to 1.25.

In the adjusted model predicting sleep duration on use days (Model 3), smoking more grams than one's average quantity (Level 1) was associated with shorter sleep duration from treatment days 6 to 13 (M_b = −0.27, b range −0.23 to −0.28), longer sleep duration from treatment days 30 to 36 (M_b = 0.34, b range 0.32–0.35), and was unassociated on other treatment days. Moreover, there was a significant cross-level interaction between cannabis grams and CUD severity from treatment days 1 to 5 (M_b = 0.21, b range 0.11–0.32). When probed, we found that as individuals used more grams than their own average, AYA with severe CUD had longer sleep durations (M_b = 0.33, b range = 0.02–0.70), while AYA with mild CUD had shorter sleep duration (M_b = −0.49, b range = −0.39 to −0.61), and there did not appear to be an association for AYA with moderate CUD (M_b = −0.08, b range = −0.18 to 0.05; Figure 4). Lastly, alcohol use from Days 4 to 12 (M_b = −0.64, b range = −0.57 to −0.67), weekends (M_b = 0.68, b range 0.54–0.79), baseline sleep duration (b = 0.42, p < .001), and baseline cannabis use (b = 0.01, p = .04) were positively associated with sleep duration, while cannabis grams (Level 2), CUD severity, and topiramate were not associated with sleep duration (Figure 4, Table 2).

FIGURE 4.

Time-varying adjusted effects of cannabis grams and the CUD interaction on sleep duration. Model 3 (see Method). This model was estimated using cannabis use days, with cannabis grams person-centered and cannabis use disorder (CUD) grand-mean centered. The first plot depicts the direct effects of each variable. The second plot depicts the probed interaction at mild (−1 standard deviation [SD] = 2.19 symptoms), moderate (mean = 4.35 symptoms), and severe (+1 SD = 6.51 symptoms) levels of CUD, estimated using reference values for alcohol use (no use), weekends (weekday), topiramate (placebo), and grand means for baseline sleep duration and baseline cannabis use. Black lines indicate regions where the 95% confidence interval did not contain 0/interaction effects were significant, while gray lines indicate regions where the 95% confidence interval did contain 0/interaction effects were not significant. See Table 2 for level 2 results from this model.

In the adjusted model predicting trouble sleeping on use days (Model 4), using more grams than one's average quantity (Level 1) was associated with less sleep trouble from treatment days 3 to 13 (M_IRR = 0.86, IRR range 0.85–0.88; ~14% reduction in sleep trouble for each additional cannabis gram), then was unassociated for the rest of treatment. Alcohol use was associated with less sleep trouble from treatment days 36 to the end of treatment (M_IRR = 0.20, IRR range 0.04–0.45). CUD severity (IRR = 1.10, p = .037; ~10% increase in sleep trouble per CUD symptom) and baseline sleep trouble were positively associated with sleep trouble (IRR = 1.38, p < .001), while cannabis grams (Level 2), CUD severity, the cross-level interaction, weekends, topiramate, and baseline cannabis use were unassociated with sleep trouble (Figure 5, Table 2).

FIGURE 5.

Time-varying adjusted effects of cannabis grams on trouble sleeping. Model 4 (see Method). Black lines indicate regions where the IRR 95% confidence interval did not contain 1 (significant), while gray lines indicate regions where the IRR 95% confidence interval did contain 1 (nonsignificant). See Table 2 for level 2 results from this model.

DISCUSSION

Improving treatments for cannabis use in AYA is a public health priority (Steele et al., 2020). One putative influence on cannabis treatment is sleep, as withdrawal-related sleep difficulties may reduce treatment efficacy. The current study built on prior work by examining trends in day-level associations between AYA cannabis use and sleep during MET+CBT. The results provide a temporally refined and clinically informative understanding of sleep processes during AYA cannabis treatment.

Our first hypothesis was that abstaining from cannabis on a given day would be associated with shorter sleep duration and more trouble sleeping; the strength of these associations would weaken across treatment, and these relations would be moderated by CUD severity. Findings from the present study did not support this first hypothesis. Abstaining from cannabis on a given day during treatment was unrelated to sleep duration or trouble sleeping, and these associations did not vary based on CUD severity. Typically, cannabis withdrawal-related sleep difficulties first manifest 24–48 h postcessation and peak after 2–6 days (Sullivan et al., 2022). One possibility is that AYA did not abstain from cannabis for long enough to experience alterations to sleep duration or trouble sleeping, as prior research using this sample noted the average number of consecutive days abstinent was 1.49 (SD = 2.97; Parnes et al., 2023). Considering AYA used cannabis on roughly 60% of treatment days, future work may wish to explore these associations in a treatment-seeking sample with endorsed abstinence goals.

There was mixed support for Hypothesis 2a that greater cannabis grams consumed on use days would be associated with shorter sleep duration and more trouble sleeping and that the strength of these associations would increase across treatment. When AYA smoked more grams compared to their average quantity, they had shorter sleep duration from treatment days 6 to 13 but longer sleep duration from treatment days 30 to 36. Lastly, cannabis grams were associated with less trouble sleeping from treatment days 3 to 13. These findings suggest that using more cannabis may offset (perhaps withdrawal-related) trouble sleeping early in treatment, despite self-reported shorter sleep duration, which could serve as an initial treatment barrier for AYA experiencing trouble sleeping. These findings inform time-specific considerations during MET+CBT. At or near the onset of treatment, AYA receiving CUD treatment may benefit from psychoeducation on the effects of cannabis on sleep as well as skills to manage sleep-related difficulties early in treatment related to reductions in use.

There was some support that (Hypothesis 2b) CUD severity would moderate the associations proposed in Hypothesis 2a. There was a significant interaction between CUD severity and cannabis grams from treatment days 1 to 5 such that when AYA used more grams than their average amount, those with severe CUD had longer sleep durations whereas those with mild CUD had shorter sleep duration. Additionally, CUD severity was positively related to trouble sleeping on use days across the study. These findings, coupled with results indicating poorer sleep duration during Days 6–13 when individuals engaged in greater cannabis use, suggest that early in treatment, lower levels of cannabis use may exacerbate sleep difficulties, and these effects may be more pronounced for those with severe CUD during the first week of treatment. Moreover, some of these difficulties may persist on use days for those with more severe CUD.

In models examining cannabis nonuse and sleep duration, alcohol use was associated with shorter sleep duration from Days 4 to 23 and 33 to 40. Alcohol use was also associated with shorter sleep duration from days 4 to 12 on cannabis use days. Polysubstance use, particularly cannabis and alcohol use, is highly prevalent among AYA in substance use treatment (Hayaki et al., 2016). Moreover, recent work suggests that cannabis use may attenuate the negative effects of alcohol on sleep quality (Wycoff et al., 2024). Similarly, Graupensperger et al. (2021) found AYA who use both cannabis and alcohol reported better sleep quality and fewer difficulties falling or staying asleep following cannabis use days but worse sleep quality on alcohol use days. These findings highlight the importance of assessing alcohol use among AYA seeking CUD treatment. To determine whether these alcohol findings may reflect a clinical consideration for working with AYA, replication of this finding coupled with assessing alcohol use continuously (e.g., number of standard drinks) will be an important avenue for future research.

The current several study had limitations. First, while participants in the current study were interested in receiving a cannabis intervention, they did not necessarily have abstinence goals, and co-occurring psychiatric diagnoses were exclusionary. Replication in clinical samples with abstinence goals or cooccurring psychiatric diagnoses is necessary to better understand the role of cannabis use on sleep during AYA treatment. For example, recent work suggests that cannabis use may differ as a function of underlying mental health symptoms (Walsh et al., 2024), which are highly prevalent among AYA seeking CUD treatment. Second, we were unable to determine who received sleep psychoeducation and thus were unable to control for impacts of psychoeducation on sleep. Although trouble sleeping declined across conditions in the current study, future work would benefit from examining whether the relation between cannabis use and sleep duration and trouble differs when AYA are taught specific skills to facilitate improved sleep. Similarly, the present sample was not recruited based on sleep impairments, and results may differ among AYA with sleep-related diagnoses (e.g., insomnia).

Third, although TVEM identifies specific days of treatment when associations exist, these associations may not replicate to the day across samples. TVEM associations should be associated more broadly, indicating general periods (e.g., early, middle, end) during treatment when cannabis use-sleep associations emerge (Meisel et al., 2021). Relatedly, while traditional estimates for multilevel model power indicate we are powered to detect a small within-person effect and a medium between-person effect (Arend & Schafer, 2019), research has yet to provide guidance on TVEM power estimation (Lanza & Linden-Carmichael, 2021), so we cannot be certain if our secondary analyses were sufficiently powered.

Fourth, our assessments of sleep were limited to self-report. Our indicator of sleep duration could not reflect whether AYA were asleep or awake in bed. Similarly, we did not examine or control for participant snoozing behaviors, which may be influenced by or influence sleep quality. Due to sources of error, our sleep duration times may be slightly elevated and not generalize to objective sleep measures. Moreover, our trouble sleeping item was non-specific and may have been interpreted differently across participants (e.g., long time to onset, early awakenings). Capturing sleep through actigraphy or validated measures of difficulty sleeping would provide a stronger test of the proposed hypotheses. Fifth, we were unable to determine the proximity of cannabis use to bedtimes, and thus, control for the level of intoxication at bedtime.

We used DSM-IV-TR to assess CUD, which has slight variations in CUD criteria (i.e., no criteria for legal involvement, new criteria for craving) compared to the latest DSM, DSM-V-TR (American Psychiatric Association, 2022). Our study also did not assess cannabis use motives, which may influence associations between cannabis use and sleep (e.g., sleep-related use motives). While we controlled for alcohol use, we did not examine alcohol quantity or couse of cannabis with nicotine or other substances. Future research should examine how couse impacts treatment outcomes, such as sleep. Lastly, data were collected for the present study prior to the popularization of concentrated cannabis products (e.g., “dabs,” vaporizer pens). Since nearly all use was flower at the time of the study, our reports assessed quantity in grams, did not assess use form or potency, and possibly overlooked rarer forms of use (e.g., edibles, concentrates). Given our focus on flower, results from the present study may not extend to other forms of cannabis use (e.g., concentrates, edibles) or more recently developed high potency flower products.

CONCLUSIONS

The present study examined day-level shifts in the associations between days AYA receiving MET+CBT did not engage in cannabis use and grams of cannabis use on use days with sleep duration and trouble sleeping. Trouble sleeping declined over the course of treatment, whereas sleep duration did not change. During the first 2 weeks of treatment, days AYA engaged in less cannabis use relative to their average levels were associated with shorter sleep duration and greater trouble sleeping. Across multiple models, there was also evidence that on days when AYA engaged in alcohol use, they had a shorter sleep duration and poorer sleep quality. Overall, these findings highlight the complex associations between cannabis and sleep impairments during treatment. The current study highlights the importance of CUD treatment for AYA covering sleep psychoeducation as well as skills to improve sleep, particularly early in treatment when reductions in cannabis use may exacerbate sleep impairment.

DATA AVAILABILITY STATEMENT

While our data is not publicly available, we can share our analysis code and output upon reasonable request of the first author.